G. Nikolov, M. Kuhn, A. Mcgibney, Bernd-Ludwig Wenning
{"title":"Reduced Complexity Approach for Uplink Rate Trajectory Prediction in Mobile Networks","authors":"G. Nikolov, M. Kuhn, A. Mcgibney, Bernd-Ludwig Wenning","doi":"10.1109/ISSC49989.2020.9180156","DOIUrl":null,"url":null,"abstract":"This paper presents a novel data rate prediction scheme. By combining online data rate estimation techniques with Long Short-Term Memory (LSTM) Neural Networks (NN), we are able to forecast the near future behaviour of the mobile channel. The prediction scheme is evaluated on data sets obtained from private and commercial mobile networks. By utilizing a Dense-Sparse-Dense (DSD) training in conjunction with weight rounding we reduce the size by a factor of 7.36 and complexity by 57% without any loss in accuracy of the model. Such an approach is especially attractive for low-end embedded-based hardware solutions where memory and processing power are limited.","PeriodicalId":351013,"journal":{"name":"2020 31st Irish Signals and Systems Conference (ISSC)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2020 31st Irish Signals and Systems Conference (ISSC)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ISSC49989.2020.9180156","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
This paper presents a novel data rate prediction scheme. By combining online data rate estimation techniques with Long Short-Term Memory (LSTM) Neural Networks (NN), we are able to forecast the near future behaviour of the mobile channel. The prediction scheme is evaluated on data sets obtained from private and commercial mobile networks. By utilizing a Dense-Sparse-Dense (DSD) training in conjunction with weight rounding we reduce the size by a factor of 7.36 and complexity by 57% without any loss in accuracy of the model. Such an approach is especially attractive for low-end embedded-based hardware solutions where memory and processing power are limited.